arXiv subjects
Publications and source records attributed to Jun Zhang.
The well-known Newton-Puiseux Theorem states that each real branch of a planar real analytic curve admits a Puiseux expansion. We generalize this result to characteristic orbit of an isolated singularity of a planar real analytic vector field and prove that each characteristic orbit has a `power-log' expansion.
Recent advances in large language models (LLMs) have highlighted the effectiveness of chain-of-thought reasoning in symbolic domains such as mathematics and programming. However, our study shows that directly transferring such text-based reasoning paradigms to protein function understanding is ineffective: reinforcement learning mainly amplifies superficial keyword patterns while failing to introduce new biological knowledge, resulting in limited generalization. We argue that protein function prediction is a knowledge-intensive scientific task that fundamentally relies on external biological priors and computational tools rather than purely internal reasoning. To address this gap, we propose PFUA, a tool-augmented protein reasoning agent that unifies problem decomposition, tool invocation, and grounded answer generation. Instead of relying on long unconstrained reasoning traces, PFUA integrates domain-specific tools to produce verifiable intermediate evidence. Experiments on four benchmarks demonstrate that PFUA consistently outperforms text-only reasoning models with an average performance improvement of 103%.
Based on the Ca\,II H and K lines observed by LAMOST, we employ the photospheric ($R'_{\rm HK}$) and basal ($R^+_{\rm HK,L}$) flux calibrated chromospheric activity indices to examine the relationship between chromospheric activity and the stellar rotation rate. We identify the rotation periods of 11,108 stars observed by Kepler and TESS by cross-matching our chromospheric activity catalog with previous studies. Our statistical results show that chromospheric activity increases with the rotation rate until it reaches a saturation level. As the stellar effective temperature increases from 4950 to 5850 K, the saturation values of the rotation period ($P_{\rm rot}$) vary correspondingly from 4.38 to 1.23 days for $R'_{\rm HK}$ and from 9.88 to 1.33 days for $R^+_{\rm HK,L}$. Similarly, the corresponding saturation Rossby number Ro ranges from 0.200 to 0.032 for $R'_{\rm HK}$ and from 0.302 to 0.107 for $R^+_{\rm HK,L}$. The saturation is also found to be significant in stars with thick convective zones, whereas it is less apparent in stars with higher effective temperatures. For solar-like stars in the $T_{\rm eff}$ range of 4800 to 6000 K, The values of chromospheric activity indicators are saturated when $P_{\rm rot}<1.45 $ days (Ro$<$0.100) and $P_{\rm rot}<2.85 $ days (Ro$<$0.097) for $R'_{\rm HK}$ and $R^+_{\rm HK,L}$, respectively.
We introduce xLLM, an intelligent and efficient Large Language Model (LLM) inference framework designed for high-performance, large-scale enterprise-grade serving, with deep optimizations for diverse AI accelerators. To address these challenges, xLLM builds a novel decoupled service-engine architecture. At the service layer, xLLM-Service features an intelligent scheduling module that efficiently processes multimodal requests and co-locates online and offline tasks through unified elastic scheduling to maximize cluster utilization. This module also relies on a workload-adaptive dynamic Prefill-Decode (PD) disaggregation policy and a novel Encode-Prefill-Decode (EPD) disaggregation policy designed for multimodal inputs. Furthermore, it incorporates a distributed architecture to provide global KV Cache management and robust fault-tolerant capabilities for high availability. At the engine layer, xLLM-Engine co-optimizes system and algorithm designs to fully saturate computing resources. This is achieved through comprehensive multi-layer execution pipeline optimizations, an adaptive graph mode and an xTensor memory management. xLLM-Engine also further integrates algorithmic enhancements such as optimized speculative decoding and dynamic EPLB, collectively serving to substantially boost throughput and inference efficiency. Extensive evaluations demonstrate that xLLM delivers significantly superior performance and resource efficiency. Under identical TPOT constraints, xLLM achieves throughput up to 1.7x that of MindIE and 2.2x that of vLLM-Ascend with Qwen-series models, while maintaining an average throughput of 1.7x that of MindIE with Deepseek-series models. xLLM framework is publicly available at https://github.com/jd-opensource/xllm and https://github.com/jd-opensource/xllm-service.
The precision era of multi-messenger astronomy, together with modern astrophysical, cosmological, and gravitational wave observations, increasingly points toward the existence of a ``dark" sector that cannot be explained within the framework of the Standard Model of particle physics and General Relativity. In this review, we explore extensions of standard physics and examine how observational data can be used to probe new particles and interactions. We consider a wide range of scales, from Solar System tests to galactic and cosmological observations, and investigate both conventional dark matter candidates, such as weakly interacting massive particles, and alternative scenarios including ultralight fields and primordial black holes. We discuss constraints derived from compact objects such as neutron stars, black holes, pulsars, and magnetars observations as well as from high-energy astrophysical phenomena. In addition, we analyze extensions of General Relativity involving additional scalar fields and their impact on gravitational wave signals and stochastic backgrounds from primordial black holes. We also study the capture and accumulation of dark matter in compact objects, which can alter properties such as mass, radius, and tidal deformability, and consider scenarios in which dark matter decays into Standard Model particles. While current observations already place significant limits on dark matter and modified-gravity models, upcoming experiments and observatories are expected to further probe or discover such new physics by improving constraints on particle masses and interaction strengths.
Partial discharge originates from microscopic insulation imperfections in high-voltage apparatus and is widely considered a critical marker of incipient deterioration. Conventional partial discharge detection methods are typically constrained by limited bandwidth and often rely on predefined feature extraction, which impedes reliable recognition of broadband transient signals. In this work, we employ a Rydberg atomic sensor to directly capture time-domain responses of partial discharge emissions and construct distinctive spectral fingerprints for different types. A 1D ResNet deep learning model is then applied to recognize these fingerprints from time-domain signals without manual feature engineering. Under increased source-antenna distances, where spectral features are significantly attenuated, the model attains a recognition accuracy of approximately 94\% across four partial discharge categories, demonstrating robustness to attenuation and noise. We further validate the approach in a simulated early-warning scenario, where partial discharge signals mixed with noise are analyzed and the model successfully generates predictive alarms. These results underscore the potential of integrating Rydberg-based broadband sensing with data-driven analysis for non-invasive, high-sensitivity diagnostics of electrical insulation systems.
Event cameras offer high-temporal-resolution sensing that remains reliable under high-speed motion and challenging lighting, making them promising for localization from LiDAR point clouds in GPS-denied and visually degraded environments. However, aligning sparse, asynchronous events with dense LiDAR maps is fundamentally ill-posed, as direct correspondence estimation suffers from modality gaps. We propose LEAR, a dual-task learning framework that jointly estimates edge structures and dense event-depth flow fields to bridge the sensing-modality divide. Instead of treating edges as a post-hoc aid, LEAR couples them with flow estimation through a cross-modal fusion mechanism that injects modality-invariant geometric cues into the motion representation, and an iterative refinement strategy that enforces mutual consistency between the two tasks over multiple update steps. This synergy produces edge-aware, depth-aligned flow fields that enable more robust and accurate pose recovery via Perspective-n-Point (PnP) solvers. On several popular and challenging datasets, LEAR achieves superior performance over the best prior method. The source code, trained models, and demo videos are made publicly available online.
Fine-tuning large language models (LLMs) on custom datasets has become a standard approach for adapting these models to specific domains and applications. However, recent studies have shown that such fine-tuning can lead to significant degradation in the model's safety. Existing defense methods operate at the sample level and often suffer from an unsatisfactory trade-off between safety and utility. To address this limitation, we perform a systematic token-level diagnosis of safety degradation during fine-tuning. Based on this, we propose token-level data selection for safe LLM fine-tuning (TOSS), a novel framework that quantifies the safety risk of each token by measuring the loss difference between a safety-degraded model and a utility-oriented model. This token-level granularity enables accurate identification and removal of unsafe tokens, thereby preserving valuable task-specific information. In addition, we introduce a progressive refinement strategy, TOSS-Pro, which iteratively enhances the safety-degraded model's ability to identify unsafe tokens. Extensive experiments demonstrate that our approach robustly safeguards LLMs during fine-tuning while achieving superior downstream task performance, significantly outperforming existing sample-level defense methods. Our code is available at https://github.com/Polly-LYP/TOSS.
The Distribution Matching Distillation (DMD) has been successfully applied to text-to-image diffusion models such as Stable Diffusion (SD) 1.5. However, vanilla DMD suffers from convergence difficulties on large-scale flow-based text-to-image models, such as SD 3.5 and FLUX. In this paper, we first analyze the issues when applying vanilla DMD on large-scale models. Then, to overcome the scalability challenge, we propose implicit distribution alignment (IDA) to constrain the divergence between the generator and the fake distribution. Furthermore, we propose intra-segment guidance (ISG) to relocate the timestep denoising importance from the teacher model. With IDA alone, DMD converges for SD 3.5; employing both IDA and ISG, DMD converges for SD 3.5 and FLUX.1 dev. Together with a scaled VFM-based discriminator, our final model, dubbed \textbf{SenseFlow}, achieves superior performance in distillation for both diffusion based text-to-image models such as SDXL, and flow-matching models such as SD 3.5 Large and FLUX.1 dev. The source code is available at \href{https://github.com/XingtongGe/SenseFlow}{https://github.com/XingtongGe/SenseFlow}
The integration of large language models (LLMs) into wireless networks has sparked growing interest in building autonomous AI agents for wireless tasks. However, existing approaches rely heavily on manually crafted prompts and static agentic workflows, a process that is labor-intensive, unscalable, and often suboptimal. In this paper, we propose WirelessAgent++, a framework that automates the design of agentic workflows for various wireless tasks. By treating each workflow as an executable code composed of modular operators, WirelessAgent++ casts agent design as a program search problem and solves it with a domain-adapted Monte Carlo Tree Search (MCTS) algorithm. Moreover, we establish WirelessBench, a standardized multi-dimensional benchmark suite comprising Wireless Communication Homework (WCHW), Network Slicing (WCNS), and Mobile Service Assurance (WCMSA), covering knowledge reasoning, code-augmented tool use, and multi-step decision-making. Experiments demonstrate that \wap{} autonomously discovers superior workflows, achieving test scores of $78.37\%$ (WCHW), $90.95\%$ (WCNS), and $97.07\%$ (WCMSA), with a total search cost below $\$ 5$ per task. Notably, our approach outperforms state-of-the-art prompting baselines by up to $31\%$ and general-purpose workflow optimizers by $11.1\%$, validating its effectiveness in generating robust, self-evolving wireless agents. The code is available at https://github.com/jwentong/WirelessAgent-R2.
Recent advances in 3D scene reconstruction and 4D human animation have broadened adoption, but integrating the two remains difficult. Key challenges include placing humans at plausible locations and scales without interpenetration, aligning lighting and style between humans and background, and handling dynamic camera trajectories. We introduce AnimateScene, a unified framework addressing these issues. First, an accurate placement module automatically determines realistic 3D positions and prevents collisions during motion. Second, we propose a training-free style alignment method that adapts the 4D human to match background illumination and appearance, achieving coherent composites. Finally, we design a joint post-reconstruction approach that incorporates camera trajectories for smooth, visually engaging motion videos. Experiments across diverse scenes and actions demonstrate that AnimateScene generates dynamic results with high geometric detail and strong spatiotemporal coherence. The project page is available at https://whynothaha.github.io/AnimateScene/AnimateScene.html.
Massive multiple-input multiple-output (MIMO) is a key enabler for the high data rates required by the sixth-generation networks, yet its performance hinges on effective beam management with low training overhead. This paper proposes an interpretable framework to tackle beam alignment in mixed line-of-sight (LoS) and non-line-of-sight (NLoS) propagation environments. Our approach utilizes multi-modal data to construct virtual base stations (VBSs), which are geometrically defined as mirror images of the base station across reflecting surfaces reconstructed from 3D LiDAR points. These VBSs provide a sparse and spatial representation of the dominant features of the wireless environment. Based on the constructed VBSs, we develop a VBS-assisted beam alignment scheme comprising coarse channel reconstruction followed by partial beam training. Numerical results demonstrate that the proposed method achieves near-optimal performance in terms of spectral efficiency.
Harnessing the reasoning power of Large Language Models (LLMs) for recommender systems is hindered by two fundamental challenges. First, current approaches lack a mechanism for automated, data-driven discovery of effective reasoning patterns, relying instead on brittle manual templates or unstable zero-shot prompting. Second, they employ structure-collapsing integration: direct prompting incurs prohibitive online inference costs, while feature extraction collapses reasoning chains into single vectors, discarding stepwise logic. To address these challenges, we propose SCoTER (Structured Chain-of-Thought Transfer for Enhanced Recommendation), a unified framework that treats pattern discovery and structure-aware transfer as a jointly optimized problem. Specifically, SCoTER operationalizes this through two synergistic components: a Generate-Validate-Mine (GVM) pipeline for automated pattern discovery and a structure-preserving integration architecture that transfers stepwise logic to efficient models. Empirically, experiments on four benchmarks demonstrate consistent improvements across diverse backbones. Moreover, in production deployment on the Tencent Advertising Platform, SCoTER achieved a 2.14\% lift in Gross Merchandise Value (GMV) while eliminating online LLM inference costs. Overall, SCoTER presents a practical and unified framework for integrating structured LLM reasoning into recommender systems, validated by consistent improvements in both offline benchmarks and online production environments.
Generative recommendation models sequence generation to produce items end-to-end, but training from behavioral logs often provides weak supervision on underlying user intent. Although Large Language Models (LLMs) offer rich semantic priors that could supply such supervision, direct adoption in industrial recommendation is hindered by two obstacles: semantic signals can conflict with platform business objectives, and LLM inference is prohibitively expensive at scale. This paper presents S-GRec, a semantic-aware framework that decouples an online lightweight generator from an offline LLM-based semantic judge for train-time supervision. S-GRec introduces a two-stage Personalized Semantic Judge (PSJ) that produces interpretable aspect evidence and learns user-conditional aggregation from pairwise feedback, yielding stable semantic rewards. To prevent semantic supervision from deviating from business goals, Asymmetric Advantage Policy Optimization (A2PO) anchors optimization on business rewards (e.g., eCPM) and injects semantic advantages only when they are consistent. Extensive experiments on public benchmarks and a large-scale production system validate both effectiveness and scalability, including statistically significant gains in CTR and a 1.19\% lift in GMV in online A/B tests, without requiring real-time LLM inference.
Multivariate time-series anomaly detection is essential for reliable industrial control, telemetry, and service monitoring. However, the evolving inter-variable dependencies and inevitable noise render it challenging. Existing methods often use single-scale graphs or instance-level contrast. Moreover, learned dynamic graphs can overfit noise without a stable anchor, causing false alarms or misses. To address these challenges, we propose the CGSTA framework with two key innovations. First, Dynamic Layered Graph Construction (DLGC) forms local, regional, and global views of variable relations for each sliding window; rather than contrasting whole windows, Contrastive Discrimination across Scales (CDS) contrasts graph representations within each view and aligns the same window across views to make learning structure-aware. Second, Stability-Aware Alignment (SAA) maintains a per-scale stable reference learned from normal data and guides the current window's fast-changing graphs toward it to suppress noise. We fuse the multi-scale and temporal features and use a conditional density estimator to produce per-time-step anomaly scores. Across four benchmarks, CGSTA delivers optimal performance on PSM and WADI, and is comparable to the baseline methods on SWaT and SMAP.
Video diffusion models (DMs) have enabled high-quality video synthesis. However, their computation costs scale quadratically with sequence length because self-attention has quadratic complexity. While linear attention lowers the cost, fully replacing quadratic attention requires expensive pretraining due to the limited expressiveness of linear attention and the complexity of spatiotemporal modeling in video generation. In this paper, we present LinVideo, an efficient data-free post-training framework that replaces a target number of self-attention modules with linear attention while preserving the original model's performance. First, we observe a significant disparity in the replaceability of different layers. Instead of manual or heuristic choices, we frame layer selection as a binary classification problem and propose selective transfer, which automatically and progressively converts layers to linear attention with minimal performance impact. Additionally, to overcome the ineffectiveness and inefficiency of existing objectives for this transfer process, we introduce an anytime distribution matching (ADM) objective that aligns the distributions of samples across any timestep along the sampling trajectory. This objective is efficient and recovers model performance. Extensive experiments show that our method achieves a 1.25-2.00x speedup while preserving generation quality, and our 4-step distilled model further delivers a 15.92x latency reduction with minimal visual quality drop.
Mixture-of-Experts (MoE) Multimodal large language models (MLLMs) excel at vision-language tasks, but they suffer from high computational inefficiency. To reduce inference overhead, expert skipping methods have been proposed to deactivate redundant experts based on the current input tokens. However, we find that applying these methods-originally designed for unimodal large language models (LLMs)-to MLLMs results in considerable performance degradation. This is primarily because such methods fail to account for the heterogeneous contributions of experts across MoE layers and modality-specific behaviors of tokens within these layers. Motivated by these findings, we propose MoDES, the first training-free framework that adaptively skips experts to enable efficient and accurate MoE MLLM inference. It incorporates a globally-modulated local gating (GMLG) mechanism that integrates global layer-wise importance into local routing probabilities to accurately estimate per-token expert importance. A dual-modality thresholding (DMT) method is then applied, which processes tokens from each modality separately, to derive the skipping schedule. To set the optimal thresholds, we introduce a frontier search algorithm that exploits monotonicity properties, cutting convergence time from several days to a few hours. Extensive experiments for 3 model series across 13 benchmarks demonstrate that MoDES far outperforms previous approaches. For instance, when skipping 88% experts for Qwen3-VL-MoE-30B-A3B-Instruct, the performance boost is up to 10.67% (97.33% vs. 86.66%). Furthermore, MoDES significantly enhances inference speed, improving the prefilling time by 2.16$\times$ and the decoding time by 1.26$\times$. Our code is available at https://github.com/ModelTC/MoDES.
Reasoning distillation has attracted increasing attention. It typically leverages a large teacher model to generate reasoning paths, which are then used to fine-tune a student model so that it mimics the teacher's behavior in training contexts. However, previous approaches have lacked a detailed analysis of the origins of the distilled model's capabilities. It remains unclear whether the student can maintain consistent behaviors with the teacher in novel test-time contexts, or whether it regresses to its original output patterns, raising concerns about the generalization of distillation models. To analyse this question, we introduce a cross-model Reasoning Distillation Provenance Tracing framework. For each action (e.g., a sentence) produced by the distilled model, we obtain the predictive probabilities assigned by the teacher, the original student, and the distilled model under the same context. By comparing these probabilities, we classify each action into different categories. By systematically disentangling the provenance of each action, we experimentally demonstrate that, in test-time contexts, the distilled model can indeed generate teacher-originated actions, which correlate with and plausibly explain observed performance on distilled model. Building on this analysis, we further propose a teacher-guided data selection method. Unlike prior approach that rely on heuristics, our method directly compares teacher-student divergences on the training data, providing a principled selection criterion. We validate the effectiveness of our approach across multiple representative teacher models and diverse student models. The results highlight the utility of our provenance-tracing framework and underscore its promise for reasoning distillation. We hope to share Reasoning Distillation Provenance Tracing and our insights into reasoning distillation with the community.